Using pre-trained models to detect objects in an image. The code is based on the jupyter notebook in this repository
import numpy as np
import os
import tensorflow as tf
import time
from matplotlib import pyplot as plt
from PIL import Image
from glob import glob
%matplotlib inline
from utils import label_map_util
from utils import visualization_utils as vis_util
# Size, in inches, of the output images.
IMAGE_SIZE = (8, 5)
#get image paths from directory
def get_image_paths(PATH_TO_TEST_IMAGES_DIR):
print(os.path.join(PATH_TO_TEST_IMAGES_DIR, '*.jpg'))
image_paths = glob(os.path.join(PATH_TO_TEST_IMAGES_DIR, '*.jpg'))
print("Length of test images:", len(image_paths))
return image_paths
def load_image_into_numpy_array(image):
(im_width, im_height) = image.size
return np.array(image.getdata()).reshape(
(im_height, im_width, 3)).astype(np.uint8)
def load_graph(model_path):
detection_graph = tf.Graph()
with detection_graph.as_default():
od_graph_def = tf.GraphDef()
with tf.gfile.GFile(model_path, 'rb') as fid:
serialized_graph = fid.read()
od_graph_def.ParseFromString(serialized_graph)
tf.import_graph_def(od_graph_def, name='')
return detection_graph
def run_detection(detection_graph, image_paths, fx, fy, img_width, img_height):
with detection_graph.as_default():
with tf.Session(graph=detection_graph) as sess:
# Definite input and output Tensors for detection_graph
image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
# Each box represents a part of the image where a particular object was detected.
detection_boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
# Each score represent how level of confidence for each of the objects.
# Score is shown on the result image, together with the class label.
detection_scores = detection_graph.get_tensor_by_name('detection_scores:0')
detection_classes = detection_graph.get_tensor_by_name('detection_classes:0')
num_detections = detection_graph.get_tensor_by_name('num_detections:0')
for image_path in image_paths:
image = Image.open(image_path)
# the array based representation of the image will be used later in order to prepare the
# result image with boxes and labels on it.
image_np = load_image_into_numpy_array(image)
# Expand dimensions since the model expects images to have shape: [1, None, None, 3]
image_np_expanded = np.expand_dims(image_np, axis=0)
time0 = time.time()
# Actual detection.
(boxes, scores, classes, num) = sess.run(
[detection_boxes, detection_scores, detection_classes, num_detections],
feed_dict={image_tensor: image_np_expanded})
time1 = time.time()
boxes = np.squeeze(boxes)
scores = np.squeeze(scores)
classes = np.squeeze(classes).astype(np.int32)
# Visualization of the results of a detection.
vis_util.visualize_boxes_and_labels_on_image_array(
image_np, boxes, classes, scores,
category_index,
use_normalized_coordinates=True,
line_thickness=6)
plt.figure(figsize=IMAGE_SIZE)
plt.imshow(image_np)
plt.show()
print(image_path,"\n")
min_score_thresh = .50
for i in range(boxes.shape[0]):
if scores is None or scores[i] > min_score_thresh:
class_name = category_index[classes[i]]['name']
print('{}'.format(class_name), scores[i])
perceived_width_x = (boxes[i][3] - boxes[i][1]) * img_width
perceived_width_y = (boxes[i][2] - boxes[i][0]) * img_height
# ymin, xmin, ymax, xmax = box
# depth_prime = (width_real * focal) / perceived_width
perceived_depth_x = ((.1 * fx) / perceived_width_x)
perceived_depth_y = ((.3 * fy) / perceived_width_y )
estimated_distance = round((perceived_depth_x + perceived_depth_y) / 2)
print("Distance (metres)", estimated_distance)
print("Time in milliseconds", (time1 - time0) * 1000, "\n")
TEST_IMAGE_PATHS_SIM = get_image_paths('test_images_sim')
TEST_IMAGE_PATHS_REAL = get_image_paths('test_images_udacity')
ssd_mobile_sim_model = 'frozen_sim_mobile-5000/frozen_inference_graph.pb'
ssd_mobile_real_model = 'frozen_real_mobile-5000/frozen_inference_graph.pb'
ssd_inception_sim_model = 'frozen_sim_inception-5000/frozen_inference_graph.pb'
ssd_inception_real_model = 'frozen_real_inception-5000/frozen_inference_graph.pb'
Label maps map indices to category names, so that when our convolution network predicts 2, we know that this corresponds to Red. Here we use internal utility functions, but anything that returns a dictionary mapping integers to appropriate string labels would be fine.
PATH_TO_LABELS = 'label_map.pbtxt'
NUM_CLASSES = 4
label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
category_index = label_map_util.create_category_index(categories)
print(category_index)
detection_graph = load_graph(ssd_mobile_sim_model)
run_detection(detection_graph, TEST_IMAGE_PATHS_SIM, 1345.200806, 1353.838257, 800, 600)
detection_graph = load_graph(ssd_mobile_real_model)
run_detection(detection_graph, TEST_IMAGE_PATHS_REAL, 1345.200806, 1353.838257, 1368, 1096)
detection_graph = load_graph(ssd_inception_sim_model)
run_detection(detection_graph, TEST_IMAGE_PATHS_SIM, 1345.200806, 1353.838257, 800, 600)
detection_graph = load_graph(ssd_inception_real_model)
run_detection(detection_graph, TEST_IMAGE_PATHS_REAL, 1345.200806, 1353.838257, 1368, 1096)